CD137 Costimulation Counteracts TGFβ Inhibition of NK-cell Antitumor Function

Natural Killer (NK) cells are cytotoxic lymphocytes of the innate lymphoid cell family that also comprises type 1, 2, and 3 innate lymphoid cells (ILC1, ILC2 and ILC3), which have regulatory function (1). NK cells can directly recognize transformed cells in a neoantigen-independent manner (natural cytotoxicity) or upon tumor-cell coating with antibodies (antibody-dependent cell cytotoxicity or ADCC), a mechanism likely contributing to the clinical efficacy of tumor antigen–specific therapeutic antibodies, such as HER2-targeted antibodies (for example, trastuzumab and pertuzumab; refs. 2–4).

NK cells contribute to cancer control through their cytolytic activity and their secretion of chemokines (i.e., MIP1α, MIP1β, CCL5) and proinflammatory cytokines (i.e., IFNγ and TNFα), which recruit type I dendritic cells and other immune cells, contributing to Th1-skewed responses in the tumor and draining lymph nodes (5–7).

Despite their pivotal role in the generation of antitumor immunity, NK cells are rather scarce in solid tumors, and their function appears compromised owing to suppressive factors such as TGFβ, which inhibits NK-cell proliferation, cytolytic activity, and IFNγ production (8, 9). Moreover, TGFβ-driven conversion of NK cells into ILC1 has been proposed as a mechanism by which tumors escape surveillance by the innate immune system (10, 11). To overcome these challenges and enhance NK-cell antitumor function, it will be necessary to develop strategies for increasing their homing to the tumor and for improving their functional persistence by overcoming immunosuppression.

CD137 (also known as 4–1BB) is a surface TNFR family member originally discovered as a costimulatory receptor on antigen-primed T cells (12). CD137 costimulation promotes NK-cell proliferation and cytokine secretion, whereas its potential for enhancing NK-cell cytotoxicity remains controversial (13, 14). In the present study, we have identified CD137 as an actionable target for enhancing persistent NK-cell cytotoxicity and cytokine secretion in the context of TGFβ immunosuppression. Experiments with fresh breast carcinoma–derived multicellular cultures as surrogates of tumor microenvironment provide the proof of concept supporting CD137 as a targetable receptor for enhancing tumor-infiltrating NK-cell activation as well as CCL5 and IFNγ secretion, regardless of TGFβ imprinting.

Peripheral blood mononuclear cells (PBMC) were obtained from volunteer healthy adults (n = 30) by Ficoll-Hypaque gradient (Lymphoprep, catalog no. 1114547) and kept overnight in RPMI 1640 GlutaMAX (Invitrogen, catalog no. 72400–021) supplemented with penicillin/streptomycin (100 U/mL and 100 μg/mL, respectively, Invitrogen, catalog no. 15140–122), sodium pyruvate (1 mmol/L, Gibco, catalog no. 11360–039), 10% FBS (Gibco, catalog no. 10270–106), and recombinant IL2 (200 U/mL, Proleukin, Eurocetus). NK cells were purified by negative selection using an NK-cell isolation kit (Miltenyi, catalog no. 130–092–657) according to the manufacturer instructions.

Treatment-naive breast carcinoma tumor specimens were selected by the pathologist based on the availability of remnant material from diagnostic (n = 17, clinicopathologic features included as Supplementary Table S1), freshly collected in sterile PBS, mechanically disrupted and digested for 40 minutes with collagenase type IV (1 mg/mL, Gibco, catalog no. 17104–019) and DNAse (50×10³ U/mL, New England Biolabs, catalog no. M0303L). Multicellular suspensions were recovered and cultured in complete RPMI with IL2 (200 U/mL Proleukin, Eurocetus) and trastuzumab (210 ng/mL, Herceptin, Roche, from the hospital pharmacy) and/or urelumab (50 μg/mL, Bristol Myers Squibb, provided by the company) for 24 hours or 6 days.

All healthy volunteers and patients gave written informed consent for the analysis of peripheral blood and tumor biopsies for research purposes. Studies were conducted in accordance with Declaration of Helsinki guidelines. Study protocol was approved by the local ethics committee (Clinical Research Ethics Committee, Parc de Salut Mar n° 2015/6038/I for healthy volunteers; n° 2015/6038/I and 2019/8584/I for patients with cancer).

The human HER2-positive (HER2⁺) breast cancer cell line SKBR3 (ATCC catalog no. HTB-30, RRID:CVCL_0033) was obtained from ATCC. SKBR3 cells were grown in DMEM/F-12 medium (Sigma-Aldrich, catalog no. D6421) supplemented with L-glutamine (0.5 mmol/L, Gibco, catalog no. 25030–024). Authenticity of the cells was tested by short tandem repeat (STR) DNA profiling analysis at the ATCC (June 2013); the cells were not reauthenticated. The mouse mastocytoma cell line P815 and the HLA class I-negative human erythroleukemic cell line K562 (both kindly provided by Dr. A. Moretta in 1988, Ludwig Institute for Cancer Research) were cultured in complete RPMI 1640 GlutaMAX medium. All cell lines were tested monthly for Mycoplasma (MycoAlert Mycoplasma Detection Kit, Lonza, catalog no. LT07–418).

Purified NK cells were cultured in flat-bottom plates previously coated with purified anti-CD16 (5 μg/mL, clone KD1, produced and purified in the laboratory). TGFβ1 (10 ng/mL, Peprotech, catalog no. 100–21), the anti-CD137 agonist urelumab (50 or 1 μg/mL), or control human IgG4 (50 μg/mL, Bristol Myers Squibb) were added at 8 hours post activation. When indicated, TGFβ1 or the TGFβ-RI inhibitor SB-431542 (20 μmol/L, Sigma-Aldrich, catalog no. S4317–5) were added prior to activation with anti-CD16.

Primary or expanded purified NK cells were pretreated with aggregated human IgG (10 μg/mL, Baxter, catalog no. 812651) and stained with combinations of directly labelled antibodies. For intracellular staining, cells were treated with brefeldin A (10 μg/mL, Sigma-Aldrich, catalog no. B7651–5 mg), labeled for surface molecules prior to fixation and permeabilization (fixation/permeabilization kit, BD Biosciences, catalog no. 554714) and staining for intracellular antigens. Data was acquired on a BD LSRFortessa or BD-LSRII (BD Biosciences) and analyzed with FlowJo software (v10.0.7, Tree Star). Antibodies used for flow cytometry are listed in Supplementary Tables S2 and S3, including the mix of hybridoma supernatants used for KIR⁺ NK-cell determinations (Supplementary Table S3).

Primary or CD16-expanded NK cells were cultured in the presence of target cells (SKBR3 or K562) at 4:1 effector:target (E:T) ratio, unless stated. In ADCC assays, trastuzumab was added at 210 ng/mL. In degranulation assays, anti-CD107a-FITC (clone H4A3, BD Biosciences, catalog no. 555800) and monensin (5 μg/mL, Sigma-Aldrich, catalog no. M5273) were included. CD107a mobilization was monitored by flow cytometry after 4 hours. For cytotoxicity assays, activated caspase 3 in SKBR3 cells was analyzed by intracellular staining, gating on forward and side scatter (FSC/SSC) and CD45 exclusion by flow cytometry. Spontaneous active caspase 3 levels in SKBR3 cells were subtracted in each condition. In indicated experiments, recombinant human TNFα (10 ng/mL, Peprotech, catalog no. 300–01A), the neutralizing anti-TNFα infliximab (50 μg/mL, Remicade, Janssen, from the hospital pharmacy) or anti–TNF-RI and -RII blocking antibodies (10 μg/mL, #16805 and #22210 respectively, R&D Systems, catalog no. MAB625-SP and MAB726-SP) were added along the assay. For P815-redirected activation, purified NK cells were cocultured with P815 cells (E:T ratio 4:1) previously incubated for 20 minutes with anti-CD56 (clone C218), anti-CD16 (clone KD1), anti-NKp30 (clone AZ20), or anti-NKp46 (clone Bab281) supernatants. NK-cell degranulation and CD137 expression were analyzed by flow cytometry in 4-hour or 24-hour cocultures, respectively. Anti-NKp46 (clone Bab281), anti-NKp30 (clone AZ20) were kindly provided by Dr. A. Moretta (University of Genova, Genova).

Purified NK cells were cultured in complete RPMI supplemented with IL2 (200 U/mL) or IL15 (10 ng/mL, Preprotech, catalog no. 200–15) for 24 hours or 48 hours before being analyzed by flow cytometry. NK cells expanded for 6 days as described above (“NK-cell expansion in CD16-coated plates”) were restimulated by plate-bound anti-CD16 and IL12 (10 ng/mL, Preprotech, 200–12H) for IFNγ production analysis.

PBMCs were labeled with 0.5 μmol/L CFSE (CellTrace CFSE Cell Proliferation Kit, Invitrogen, catalog no. C34554) prior to NK-cell purification and activation by plate-bound anti-CD16. After 6 days, CFSE fluorescence was analyzed by flow cytometry and expressed as percentage of NK cells undergoing ≥4 divisions.

Gene expression microarray analysis was performed by the Mar Genomics core facility at Hospital del Mar Medical Research Institute, Barcelona, Spain. Costimulatory receptor expression was analyzed in total RNA from purified NK cells of 2 independent donors after 12 hours activation with anti-CD16 (clone KD1, 5 μg/mL). Transcriptomic changes induced by TGFβ1 and urelumab were assessed using purified NK cells of 3 independent donors after 5 days activation with anti-CD16 (clone KD1, 5 μg/mL), in the presence of IL2, TGFβ1, and urelumab (50 μg/mL), as described above (“NK-cell expansion in CD16-coated plates”). In both cases, total RNA was isolated using RNeasy Micro Kit (Qiagen, catalog no. 74004). Amplification and labelling was performed using GeneChip WT PLUS Reagent kit (P/N 703174 2017, Thermofisher, catalog no. 902280) and hybridized to either Affymetrix Human Gene 2.0 ST Array (Thermofisher, catalog no. 902112) or Human Clariom S Array (Thermo Fisher Scientific, catalog no. 902926). For analysis, R programming (Version 3.4.3) Bioconductor and the Comprehensive R Archive Network (CRAN 2017) packages were used. Samples were background corrected, quantile normalized, and summarized to a gene-level using the robust multi-chip average (RMA). An empirical Bayes moderated t-statistics model (LIMMA) was built to detect differentially expressed genes with a P value < 0.05 and a fold change > 1.4. Data represented in heatmaps was normalized in each gene using z-score normalization. Transcriptomic data from both microarrays are available in GSE156200 SuperSeries.

Differentially expressed genes between murine ILC1s and NK cells isolated from intestine and liver were generated from the Gene Expression Omnibus (GEO) dataset GSE37448 and used for PreRanked Gene Set Enrichment Analysis (GSEA). The list of differentially expressed genes between murine ILC1s and NK cells isolated from intestine and liver were obtained using GEO2R platform (15).

Production of IFNγ, TNFα, and CCL5 by NK cells during their expansion in CD16-coated plates was measured in cell-free supernatants using 88-7316-88 (eBioscience), DY210, and DY278 (R&D Systems) commercial ELISAs, respectively, following the manufacturers' instructions.

TGFβ-RII and SMAD7 were analyzed in anti-CD16 activated NK cells from healthy individuals in the presence or absence of TGFβ1 and urelumab at 24 hours post activation. NK-cell dry pellets were lysed in RIPA buffer [40 μL/10⁶ cells; 0.15M NaCl, Sigma, catalog no. S9888; 1% Nonidet P-40, GmbH, catalog no. 1332473; 0.1% SDS, Sigma, catalog no. L-3771; 50 mmol/L Tris, pH 7.5, Sigma, catalog no. T-8404, supplemented with protease and phosphatase inhibitors: 1 mmol/L phenylmethylsulfonylfluoride (PMSF), Sigma, catalog no. P7626; 2 mmol/L Na₃VO₄, Sigma, catalog no. S-6508; 5 mmol/L NaF, Merck, catalog no. 106449; 10 mmol/L β-glycerol-phosphate, Sigma, catalog no. G9422;, 5 mmol/L EDTA, Sigma, catalog no. E5134; and 1 mmol/L protease inhibitor cocktail Sigma, Sigma-Aldrich, catalog no. P8340]. Protein extracts corresponding to 0.5 × 10⁶ NK cells were denatured in Laemmli buffer (Biorad, catalog no. 161–0747), boiled 5 minutes at 95 °C, separated by 10% SDS-PAGE under reducing conditions, and transferred onto polyvinylidene difluoride (PVDF) membranes (Immobilon, Millipore, catalog no. IPVH00010). Membranes were blocked in 5% BSA or nonfat dry milk (Asturiana, catalog no. AAA022) in TBS + 0.05% Tween 20 (Sigma, catalog no. P7949) and blotted with specific antibodies: anti-SMAD7 (catalog no. MAB2029, R&D Systems, 1:1000), anti–TGFβ-RII (catalog no. sc-17791, Santa Cruz Biotechnology, 1:100) and anti–β-actin (Sigma-Aldrich, catalog no. A5441, 1:15000). Proteins were detected with HRP-labeled anti-rabbit or anti-mouse (GE Healthcare, catalog no. NA934 and NA931V respectively, 1:3000 except for β-actin 1:10000) and enhanced chemiluminescence (ECL) substrate (SuperSignal West Pico Chemiluminescent Substrate, Thermofisher, catalog no. 34080). Band intensity was quantified by ImageJ (NIH).

Phospho-SMAD2 and CD56 were analyzed on consecutive sections of pretreatment formalin-fixed, paraffin-embedded (FFPE) tumor biopsies (n = 4) by IHC staining. Mouse monoclonal anti-CD56 (clone 123–3, Dako-Agilent, catalog no. M730429–2) and rabbit monoclonal anti–phospho-Smad2 (Ser465/467; 138D4, Cell Signaling Technology, catalog no. 3108S) were used, followed by incubation with a polymer coupled with peroxidase (Envision Flex, Agilent, catalog no. GV800). Sections were visualized with 3,30-diaminobenzidine and counterstained with hematoxylin. All incubations were performed using the Autostainer Link 48 platform (Agilent). Slides were scanned at 20× with an Aperio CS2 scanner and analyzed with QuPath (16) for the quantification of regions of interest (ROI). For each sample, 5 ROI equivalent to an area of 1 high-power field (HPF; 40×) each, were analyzed by applying cell Detection, object classification (identifying tumor, stroma, and immune cells), and staining intensity. Categories for staining intensity classification included: 0 (no staining), 1+ (weak staining), 2+ (moderate staining), 3+ (strong staining). An H-score was calculated for each tissue classification by combining staining intensity and the percentage of cells in each staining class, with a maximum score of 300. Scoring was performed by an expert pathologist (L. Comerma, Institut Hospital del Mar d'Investigacions Mèdiques, Barcelona, Spain).

Gene expression microarray data of treatment-naive HER2⁺ breast tumors from patients with breast cancer undergoing trastuzumab-based neoadjuvant treatment in the context of a phase II randomized clinical trial was downloaded from GEO (https://www.ncbi.nlm.nih.gov/geo/; GSE130786 dataset). Agilent array was quantile normalized and log₂ transformed. All data was processed within R. Gene set variation analysis (GSVA) of the NK gene signature (CD160, CD244, CTSW, FASLG, GZMA, GZMB, GZMH, IL18RAP, IL2RB, KIR2DL4, KLRB1, KLRC3, KLRD1, KLRF1, KLRK1, NCR1, NKG7, PRF1, XCL1, and XCL2 genes; ref. 17) was computed with the GSVA R package. Z scores for GZMB, IFNG, CCL5, TNFRSF9, TNFSF9, KLRK1, CXCL9, CXCL10, TGFB1, TGFB2, TGFB3, and FCGR3A transcripts were used for analyzing their association with pathologic complete response (pCR) to trastuzumab treatment. Odds ratio for pCR achievement were calculated for the top and bottom quartile expression ranked values for selected genes or the ranked sum expression of the NK gene signature.

Unless stated, statistical significance was calculated using paired or unpaired Student t test for nonnormalized or normalized data, respectively (*P < 0 0.05; **P < 0 0.01; ***P < 0.001; ****P < 0.0001).

To identify targetable mechanisms enhancing NK cell–mediated ADCC we analyzed changes in costimulatory and coinhibitory receptor expression in primary NK cells activated via CD16. Gene expression microarray data showed that cross-linking of CD16 promoted the transcription of several inducible members of the TNFRSF, among which, TNFRSF9 (CD137, 4–1BB) displayed the largest induction as compared with TNFRSF18 (GITR), TNFRSF8 (CD30), and TNFRSF4 (OX40; Fig. 1A). Similar patterns of induction were seen at the protein level in trastuzumab-induced ADCC assays against the HER2⁺ breast cancer cell line SKBR3 (Fig. 1B and C; Supplementary Fig. S1). Expression of CTLA4 and PDCD1 (which encodes PD1) mRNA was not induced in CD16-activated NK cells, nor was PD1 protein expression detected in NK cells upon trastuzumab-induced ADCC (Fig. 1A–C; Supplementary Fig. S1). Surface expression of CD137 upon trastuzumab-induced activation was mostly detected in CD56^dim NK cells, displaying earlier kinetics, larger magnitude, and sustained persistence, as compared with OX40 (Supplementary Fig. S1). NK-cell stimulation by IL2 and IL15 or by redirected activation through distinct NK-cell receptors (NKR) coupled to tyrosine kinase–mediated signaling pathways (i.e., CD16, NKp30, NKp46) induced surface CD137 upregulation (Supplementary Fig. S2), in agreement with prior reports (18, 19). The frequency of NK cells upregulating CD137 was proportional to the percentage of NK cells degranulating upon redirected activation (Supplementary Fig. S2).

Based on the prominent production of TNFα during NK cell–mediated ADCC (Supplementary Fig. S1), we analyzed whether TNFα contributed to CD137 expression in NK cells, as has been previously described in regulatory T cells (20). Addition of the neutralizing anti-TNFα infliximab or recombinant TNFα to ADCC assays decreased and enhanced CD137 expression in NK cells, respectively (Fig. 1D). Trastuzumab-induced CD137 expression was evident as early as 4 hours post ADCC, whereas the reduction of CD137 levels by TNFα blockade was only evident at 24 hours post ADCC (Fig. 1E), indicating a role for TNFα in regulating CD137 persistence upon its induction by CD16 ligation. TNFα signals through either TNF-RI or TNF-RII. Expression of TNF-RI and TNF-RII was detected in resting NK cells, but only TNF-RII (TNFRSF1B) expression was enhanced upon activation via CD16 (Fig. 1A and F). Simultaneous blockade of both TNF receptors by specific antibodies significantly inhibited CD137 expression, whereas single blockade of either TNF-RI or TNF-RII resulted in a partial inhibition (Fig. 1G).

Overall, CD137 expression in NK cells was triggered by activating receptors coupled to tyrosine-kinase signaling pathways and sustained by autocrine TNFα signaling through the coordinated action of TNF-RI and TNF-RII.

We next evaluated whether costimulation via CD137 could enhance NK-cell activity in the presence of TGFβ, a major immunosuppressive factor in the tumor microenvironment (21). CFSE-labeled purified NK cells were activated via CD16 in the presence of IL2 and subsequently treated with TGFβ1 and the agonistic anti-CD137 urelumab (22). Activation via CD16 triggered vigorous NK-cell proliferation in response to IL2, which was almost completely abolished by TGFβ1 (Fig. 2A and B). CD137 cross-linking prevented TGFβ1 inhibition of CD16-induced NK-cell proliferation in a dose-dependent manner, restoring NK-cell numbers, in addition to enhancing basal NK-cell proliferation (Fig. 2A–C). Indeed, CD137 costimulation preserved the upregulation of CD25 induced by CD16 in NK cells, at 36 hours and 6 days of culture, bypassing the inhibition of TGFβ1 and enabling their proliferation in response to IL2 (Fig. 2D and E). Combined staining for surface CD56 and CD117 (cKIT) showed that the number of CD56^bright NK cells (identified as CD56^brightCD117^bright cells) remained relatively stable in all conditions, whereas the number of CD56^dim NK cells decreased in the presence of TGFβ1 unless including the CD137 agonist (Supplementary Fig. S3). Urelumab treatment also rescued CD25 expression and cell recovery when NK cells were preincubated with TGFβ1 prior to CD16 cross-linking, despite the reduced upregulation of surface CD137 in this setting (Supplementary Fig. S4).

To characterize global changes in the transcriptome of CD16-activated NK cells cultured in the presence of TGFβ1 and urelumab, we performed microarray analysis. Based on microarray data, 199 genes were differentially expressed in urelumab-treated versus nontreated CD16-activated NK cells in the presence of TGFβ1 (Fig. 2F; Supplementary Table S4). Unsupervised Ingenuity Pathway Analysis (IPA) identified “cellular growth and proliferation,” “cell death and survival,” “cell interaction,” “cell cycle,” and “immune cell trafficking” as the biological functions enriched in CD137-costimulated NK cells, regardless of TGFβ1 inhibition (Fig. 2G). Analysis of upstream regulators of differentially expressed genes predicted the activation of TNF (P value 0.003, z score 2.5) and NF-κB complex (P value ≤ 0.001, z score 2.2). Indeed, CD137 costimulation increased the expression of genes involved in: (i) CD137 signaling (i.e., TRAF1, TAB1, NFKBIA, NFKBIB, NFKB1, NFKB2); (ii) NK-cell proliferation (i.e., MYC, CCNB1, CCNB2, CDK4 or CDK6), and (iii) IL2 receptor transcripts (IL2RA, IL2RB) together with antiapoptotic molecules (i.e., BCL2A1; Fig. 2H–J).

GSEA of gene expression microarray data from NK cells cultured with TGFβ1 with or without urelumab for 5 days compared with liver and intestine ILC1 (GSE37448) and IL2- and IL15-stimulated NK cells (GSE22886) showed that NK cells activated in the presence of TGFβ1 presented a gene expression profile reminiscent to that of liver- and intestine-isolated ILC1, including the upregulation of signature genes such as ZNF683 (HOBIT) and ITGAE (CD103; Fig. 3A and B; refs. 10, 23, 24). In contrast, CD137-costimulated NK cells showed a gene expression profile reminiscent of IL2- and IL15-activated NK cells, despite the presence of TGFβ1 (Fig. 3A). In fact, CD137 costimulation partially prevented TGFβ1-downregulation of transcript levels for activating receptors, cytotoxic effector molecules, and proinflammatory cytokines (CD226, GZMB, IFNG, TNFA) and spared or further enhanced the tissue-residency gene expression program triggered by TGFβ1 [HOBIT (ZNF683), CD103 (ITGAE), and CXCR3; Fig. 3B]. Partial recovery of NKG2D and granzyme-B expression was confirmed by multiparametric flow cytometry in NK cells after 6 days of culture in CD16-coated plates in the presence of TGFβ1 and urelumab (Fig. 3C). CD16 levels were comparably reduced in all treatments (Fig. 3C), whereas no changes in surface NKp46 expression were detected (Supplementary Fig. S5A). In agreement with mRNA data, CD137 costimulation did not prevent TGFβ1-triggered expression of CD103 and CXCR3 and enhanced CCR7 expression in a proportion of NK cells (Fig. 3C). The impact of TGFβ1 and CD137 costimulation was comparable in distinct CD56^dim NK-cell subsets, defined by the expression of NKG2A in the absence of KIR or as KIR⁺NKG2A^+/– (Fig. 3D; ref. 25).

CD137 ligation in freshly isolated NK cells had no effect on basal antibody-dependent cell-mediated cytotoxicity(ADCC) induced by trastuzumab against SKBR3 cells (Supplementary Fig. S5B). However, after 6 days of CD137 costimulation in the presence of TGFβ1, NK cells displayed a preserved capacity to directly recognize and degranulate against K562 cells (Fig. 4A and B). Treatment with urelumab also prevented TGFβ1 inhibition of NK-cell cytotoxicity upon direct and trastuzumab-dependent recognition of SKBR3 cells (Fig. 4C–E; Supplementary Fig. S5C and S5D). In addition, CD137 costimulation also rescued, to a variable extent, the secretion of CCL5 (RANTES), IFNγ, and TNFα by CD16-activated NK cells in the presence of TGFβ1 (Fig. 4F–H) as well as their capacity to secrete IFNγ in response to IL12 upon restimulation (Supplementary Fig. S5E). As compared with TGFβ1-treated, CD137-costimulated NK cell cultures showed higher proportions of IFNγ⁺ and IFNγ⁺GzmB⁺ cells coexpressing CCR7 and CXCR3 (Fig. 4I–L). Overall, CD137 costimulation counteracted TGFβ inhibition, preserving NK cell capacity for direct and antibody-dependent recognition of tumor targets, CCL5, TNFα, and IFNγ production. In addition, TGFβ1 and CD137 combined signaling promoted the expression of tumor and lymph-node homing molecules in NK cells such as CXCR3 and CCR7.

We next sought to investigate the mechanisms underlying the cross-talk between CD137 ligation and TGFβ1 in CD16-activated NK cells. SB-431542, a specific inhibitor of TGFβ-RI kinase activity (26) was included during NK-cell activation via CD16 in the presence of TGFβ1 and urelumab. Addition of SB-431542 prevented TGFβ1 suppression of NK-cell proliferation and CD25 expression to a comparable extent to that of CD137 cross-linking (Fig. 5A). Induction of CD103 by TGFβ1 was also dependent on TGFβ-RI kinase activity, whereas downregulation of NKG2D and granzyme B was only partially prevented by SB-431542 (Fig. 5A). As previously suggested in phenotyping experiments, both TGFβ-RI and CD137 combined signaling were necessary for CCR7 expression (Fig. 5A).

CD137 cross-linking activates NF-κB, which has been described to modulate TGFβ canonical signaling by either promoting expression of the inhibitory SMAD7 or decreasing expression of TGFβ-RII (9, 27). A 24-hour treatment with urelumab limited upregulation of TGFβ-RII induced by TGFβ1 and slightly increased SMAD7 levels in CD16-activated NK cells (Fig. 5B and C). These results indicated that CD137 costimulation dampened the transcriptional program downstream of TGFβ-RI by reducing expression of TGF-β-RII while maintaining SMAD7 levels in activated NK cells.

To further investigate the potential of targeting CD137 in tumor-infiltrating NK cells, multicellular tumor cultures were generated by mechanical and enzymatic dissociation of treatment-naïve human breast carcinoma specimens. Overnight treatment of breast carcinoma–derived multicellular cultures with trastuzumab induced the expression of CD137 on tumor-infiltrating CD16⁺ NK cells. CD137 acquisition was associated with reduced CD16 surface levels, indicative of CD16-dependent NK-cell activation (Fig. 6A–C; Supplementary Fig. S6). No changes in CD137 expression were detected on CD8⁺ and CD4⁺ T cells concomitantly present (Fig. 6D), suggesting that NK cells are the tumor-infiltrating lymphocytes activated in response to trastuzumab. To ascertain TGFβ activity in breast carcinomas used in ex vivo experiments, phosphorylated-SMAD2 (pSMAD2) was analyzed by IHC in some of the paired FFPE tumor blocks. Nuclear pSMAD2 was detected in all 4 samples analyzed, mainly concentrated in fibroblasts and immune cells in the tumor stroma (Fig. 6E–H). We observed that pSMAD2⁺ fibroblastic cells often separated immune-rich stroma from tumor cell islands. Analysis of CD56 and pSMAD2 in consecutive FFPE sections showed active TGFβ signaling in CD56⁺ tumor-infiltrating immune cells (Fig. 6F). The average percentage of pSMAD2⁺ immune cells in breast carcinomas was similar to the proportions of CD103⁺ tumor-infiltrating NK cells, as analyzed by flow cytometry in fresh tumor samples (Fig. 6I). Based on these data supporting that tumor-infiltrating NK cells are exposed to TGFβ in breast carcinomas, we next assessed the influence trastuzumab and urelumab in breast carcinoma–derived multicellular cultures. Combined treatment with trastuzumab and urelumab enhanced the proportions of NK cells recovered at 6 days in several tumor-derived multicellular cultures as compared with those treated with trastuzumab alone (Fig. 6J and K). Higher proportions of CD103⁺CD16⁺ NK cells were detected in cultures treated with trastuzumab or trastuzumab and urelumab combination, suggestive of TGFβ signaling during activation (Fig. 6L). At 6 days, enhanced CCL5 and IFNγ secretion was detected in cell-free supernatants from cultures treated with the trastuzumab and urelumab combination (Fig. 6M–P), as an indirect assessment of tumor-infiltrating NK-cell function.

We have previously shown that numbers of tumor-infiltrating NK cells are associated with pCR to anti-HER2-based treatment in patients with primary breast cancer (2, 28). To indirectly address the putative contribution of tumor-infiltrating NK cells, the CD137 axis, and TGFβ levels to trastuzumab therapeutic activity, we analyzed a publicly available gene expression dataset from HER2⁺ breast carcinomas generated in the context of a phase II randomized trial (GSE130786). The average NK-cell score in breast carcinomas from patients achieving pCR (n = 40) with trastuzumab was higher as compared with that in those patients having partial or no response to treatment (n = 41; Fig. 7A). NK-cell score positively and significantly correlated with GZMB, CCL5, and IFNG levels, supporting a direct relationship between tumor-infiltrating NK cells and tumor microenvironments with cytotoxic and IFNγ-producing potential (Fig. 7B). No correlation between tumor-infiltrating NK-cell enrichment score and TGFβ levels was detected (Fig. 7B). TNFRSF9 (CD137) levels highly correlated with tumor-infiltrating NK-cell scores as well as with IFNG, GZMB, CCL5, and FCGR3A (CD16) levels in breast carcinomas (Fig. 7C and D). TNFSF9 (CD137L) levels were low in most tumors and moderately correlated with TNFRSF9 (Fig. 7D and E). Levels of constitutively expressed [i.e., CD16 (FCGR3A) and NKG2D (KLRK1)] or activation-induced [i.e., CD137 (TNFRSF9)] receptors with NK-cell activating/costimulatory function as well as IFNγ (IFNG) and IFNγ-induced CXCL9/10 were higher in breast carcinomas achieving pCR to treatment in univariate analysis (Fig. 7E). Among those genes, multivariate logistic regression including the NK-cell signature, CCL5, IFNG, GZMB, TGFB1, TGFB2, and TGFB3 levels pointed to IFNG [odds ratio (OR) 35.2, 2.5–97.5% confidence interval (CI) 4.6–859.5, P = 0.004] as the driver of the tumor-infiltrating NK-cell association with pCR to trastuzumab and TGFB2 levels as a negative predictor of pCR (OR 0.27, 2.5–97.5% CI, 0.07–0.89, P = 0.04; Fig. 7F).

Altogether, these results support CD137 as a targetable axis for enhancing tumor-infiltrating NK-cell persistence, IFNγ secretion and the efficacy of HER2-targeted antibodies in HER2⁺ breast cancer.

TGFβ is an important immunosuppressive factor that inhibits tumor-infiltration by immune cells and suppresses NK-cell activity (8, 21, 29). Hence, strategies circumventing TGFβ suppression in the tumor microenvironment would be of great value for rewiring NK-cell antitumor function and broadening immunotherapy efficacy. Our study shows that costimulation through CD137 enhances NK-cell proliferation and function in the context of TGFβ immunosuppression, contributing to their differentiation towards tissue-resident cytotoxic lymphocytes with preserved CCL5 and IFNγ secretion. As ascertained in human breast carcinoma samples, trastuzumab treatment induced the upregulation of CD137 in CD16⁺ breast carcinoma–infiltrating NK cells, enabling the agonistic action of CD137 agonists.

An unbiased analysis of activation-induced receptor expression in human primary NK cells pointed to CD137 as the receptor showing the highest expression upon in vitro ADCC. This is reminiscent of the induction of CD137 observed in peripheral-blood NK cells from patients with cancer right after receiving systemic treatment with tumor antigen–specific therapeutic antibodies such as cetuximab or rituximab (13, 19, 30). Expression of CD137 was induced by CD16 activation and sustained by autocrine/paracrine TNFα signaling, in agreement with studies showing transcriptional regulation of the TNFRSF9 (CD137) promoter by NF-κB and AP-1 transcription factors in T cells (31). CD137 costimulation partially rescued TNFα secretion in TGFβ-treated NK cells, likely generating a positive feedback loop contributing to reduce TGFβ inhibition.

Our study also uncovered the potential of CD137 costimulation for subverting TGFβ-induced conversion of NK cells from an ILC1-like (10, 11) back to an NK cell–like transcriptomic and functional profile. The combined action of CD137 and TGFβ signaling promoted the differentiation of NK cells with tissue-resident features (based on the expression of HOBIT and the αE integrin CD103), which partially preserved their capacity for recognizing and killing tumor cells. Preserved expression of receptors involved in tumor-cell recognition (NKG2D, NKp46) and cytolytic mediators (Granzyme B) likely accounted for their persistent cytotoxicity upon rechallenge with tumor cells.

CD137 costimulated NK cells also preserved their regulatory function despite TGFβ signaling and acquired tumor- and lymph node–homing potential (i.e., CXCR3, CCR7 expression; ref. 32), reminiscent of the helper profile induced by IL18 in human NK cells (33). In some models, the generation of T-cell antitumor immunity by the administration of a CD137 agonistic mAb is NK cell–dependent (34, 35). According to our results, CD137-dependent differentiation of cytotoxic, IFNγ-producing CXCR3⁺ and CCR7⁺ NK cells could simultaneously skew tumor and draining lymph-node microenvironments for efficiently boosting antitumor adaptive immunity. CCR7 expression in NK cells was enhanced by simultaneous CD137 and TGFβ canonical signaling. Enhancement of NF-κB1 and JunB expression by CD137 and TGFβ combined signaling could directly induce CCR7 expression (36).

The comparison of the effects of SB-431542 and urelumab showed that CD137 signaling almost completely recovered the expression of those targets exclusively dependent on TGFβ-RI canonical signaling (i.e., NK-cell proliferation, myc, CD25 expression, and CCL5), whereas the recovery of those molecules that showed partial dependency on TGFβ-RI was incomplete (i.e., GzmB, NKG2D, and IFNγ). In addition, the fact that CD137 costimulation did not prevent the acquisition of TGFβ-induced tissue-residence features (i.e., CXCR3, CD103, and Hobit), support the existence of at least two TGFβ canonical-signaling branches in activated NK cells that regulate the inhibition of proliferation/NK-cell effector activity and the acquisition of a tissue-residency profile, respectively, in analogy to the branching described in hematopoietic stem cells (37). Indeed, CD137 costimulation prevented TGFβ inhibition of those NK-cell molecules/functions shown to be regulated by SMAD3- and SMAD4-dependent signaling (NKG2D, granzyme B, IFNγ expression, and NK-cell proliferation; refs. 10, 29, 38, 39), whereas TGFβ-RI–dependent SMAD4-independent targets (i.e., CD103 and CCR7) were not affected by CD137 costimulation (10, 38–40). The reduction of TGFβ–RII expression together with the induction of the inhibitory SMAD7 in CD137 costimulated NK cells, could lead to decreased TGFβ canonical signaling, perhaps facilitating the rescue of several SMAD4-inhibited transcriptional targets (i.e., IFNγ, granzyme B, CCL5, CD25, c-Myc) by CD137-dependent activation of NF-κB (41–45), as described for proinflammatory cytokines (9). Deciphering the mechanisms by which CD137 costimulation and hence, NF-kB activation, interfere with TGFβ-induced SMAD4-dependent transcriptional program requires further investigation.

Our study also highlights the relevance of human tumor–derived multicellular cultures for exploring the mechanism of action of therapeutic drugs in tumor-infiltrating immune cells. In vitro treatment of fresh breast carcinoma–derived multicellular cultures with trastuzumab induced the expression of CD137 on tumor-infiltrating CD16⁺ NK cells but not on concomitantly present T lymphocytes, suggesting the involvement of NK cells in the antitumor response triggered by HER2-targeted therapeutic antibodies (2, 28). Nuclear pSMAD2 staining in fibroblastic and immune stromal cells demonstrated TGFβ signaling in treatment-naïve breast carcinomas. Trastuzumab-induced CD137 upregulation sensitized tumor-infiltrating CD16⁺ NK cells to CD137 agonists, which enhanced NK-cell persistence and the release of the immune-cell recruiting CCL5 and the proinflammatory cytokine IFNγ. In addition, bioinformatic analysis of breast carcinoma gene expression datasets provided the in vivo correlate between tumor-infiltrating NK-cell enrichment, TNFRSF9, and IFNG levels and the achievement of pCR to trastuzumab. Correlation between TNFRSF9 levels, NK-cell scores and all NK-cell effector molecules uncoupled from TNFSF9 expression, indirectly suggests that immune-cell activation in primary breast tumors is associated to CD137 expression in a context of low CD137 L availability. On the other hand, a trend for higher CXCL9 and CXCL10 levels in tumors achieving pCR to trastuzumab points to CXCR3 as an important tumor-homing/retention factor. Altogether, supporting the importance of NK-cell function and the suitability of targeting CD137 as a costimulatory axis for broadening the clinical efficacy of tumor antigen–specific antibodies.

TGFβ is normally present in the tumor microenvironment produced by cancer, stromal, and immune cells. Direct tumor-suppressive or tumor-promoting function of TGFβ on cancer cells is temporal and context dependent (46) yet it mostly behave as an immunosuppressive factor within the tumor microenvironment as well as in sentinel lymph nodes (47). Clinical development of systemic TGFβ inhibitors (small molecules, blocking antibodies) has been halted due to the pleiotropic nature of TGFβ and toxicities resulting from its role in immune-tolerance maintenance. Hence, alternative strategies targeting endogenous antagonists of TGFβ signaling may eventually be of potential therapeutic value. In this regard, the here described targeting of CD137 costimulatory axis could rescue NK-cell antitumor function from TGFβ1-mediated immunosuppression, as previously described for an IL15 superagonist/IL15Rα fusion complex (48). Currently, several strategies aimed at restricting the action of CD137 agonists to the tumor microenvironment are in clinical development, including intratumoral treatment with agonist anti-CD137 or bispecific antibodies simultaneously targeting CD137 and a tumor-associated antigen or a stromal component (49–51). These strategies could allow, in combination with NK-cell activating agents, bypassing TGFβ inhibition of NK-cell function while avoiding toxicity events associated to systemic treatment with CD137 agonists or TGFβ inhibitors (52). Results here described might have also translational implications in several immunotherapy scenarios ranging from currently ongoing clinical trials testing the combination of anti-CD137 agonists with anti–PD-1/PD-L1 blockade as well as adoptive cell therapy protocols with chimeric antigen receptor (CAR) T/NK cells comprising the CD137 costimulatory domain.

M. Cabo reports grants from Worldwide Cancer Research during the conduct of the study. S. Santana-Hernández reports personal fees from Proyecto Integrado de Excelencia - ISCIII and Fundació CELLEX during the conduct of the study. A. Rea reports grants from H2020-MSCA-ITN-2017 agreement number 765104-MATURE-NK during the conduct of the study. F. Balaguer reports other support from Elsevier, Cancer Prevention Pharmaceuticals, Norgine, and Symex outside the submitted work. L. Comerma reports personal fees and nonfinancial support from Roche outside the submitted work. P. Berraondo reports grants from Asociación Española Contra el Cáncer during the conduct of the study, as well as grants from Sanofi, Ferring, Bavarian Nordic, Hookipa, and Moderna and personal fees from MSD, Novartis, Boehringer Ingelheim, AstraZeneca, and Bristol Myers Squibb outside the submitted work. J. Albanell reports grants from ISCiii FIS during the conduct of the study, as well as personal fees from Roche, Seagen, AstraZeneca, and Daiichi Sankyo outside the submitted work. I. Melero reports grants and personal fees from Bristol Myers Squibb during the conduct of the study, as well as grants and personal fees from Roche, AstraZeneca, Bioncotech, Genmab, Pharmamar, and Alligator and personal fees from F-STAR, Gossamer, and Numab outside the submitted work. A. Muntasell reports grants from Fundación Asociación Española Contra el Cáncer, Proyecto Integrado de Excelencia ISCIII, Worldwide Cancer Research Foundation, and ISCiii/FEDER during the conduct of the study, as well as speakers bureau honoraria from Roche. No disclosures were reported by the other authors.

M. Cabo: Conceptualization, formal analysis, investigation, methodology, writing–review and editing. S. Santana-Hernández: Investigation, writing–review and editing. M. Costa-Garcia: Data curation, software, formal analysis, methodology, writing–review and editing. A. Rea: Investigation, writing–review and editing. R. Lozano-Rodríguez: Investigation. M. Ataya: Data curation. F. Balaguer: Resources. M. Juan: Resources. M.C. Ochoa: Methodology. S. Menéndez: Methodology. L. Comerma: Methodology. A. Rovira: Funding acquisition, investigation, writing–review and editing. P. Berraondo: Conceptualization. J. Albanell: Resources, funding acquisition, writing–review and editing. I. Melero: Conceptualization, resources, funding acquisition, writing–review and editing. M. López-Botet: Conceptualization, resources, funding acquisition, writing–review and editing. A. Muntasell: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft.

The authors are grateful to Marta Otero for collaborating in obtaining blood samples, Dr. Oscar Fornas for advice in flow cytometry, Drs. Lara Nonell and Magdalena Arnal for microarray data analysis, and volunteer blood donors and patients for their participation in the study. The authors are supported by coordinated research projects from Asociación Española contra el Cáncer (GCB15152947MELE) and Proyecto Integrado de Excelencia ISCIII (PIE 2015/00008); M. López-Botet, I. Melero, and A. Muntasell are supported by Worldwide Cancer Research Foundation (15–1146) and by Plan Estatal I+D Retos (SAF2016–80363-C2–1-R), Spanish Ministry of Economy and Competitiveness (MINECO, FEDER), and Generalitat de Catalunya (2017 SGR 888). A. Rea is funded by EC Horizon 2020, Marie Sklodowska Curie-Innovative Training Network (No. 765104; 2018–2021). A. Muntasell is supported by ISCiii/FEDER (PI19/PI19/00328). J. Albanell and A. Rovira are supported by ISCiii/FEDER (PI15/00146, PI18/00006, and CIBERONC) and by Generalitat de Catalunya (2017 SGR 507).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.